This is the category of technical posts that discuss the use of (finite) state machine as the primary abstraction to build sophisticated behaviours for characters in video games (and side-scrolling platformers in particular).
This is to meet the observation that, in games I loved, one cannot really speak about "artificial intelligence" for ennemies. However, they have behaviours more sophisticated than a mere goomba. More like a Zelda boss, they alternate between states where they have different hit boxes corresponding to harmful blades or weak spots. A sophisticated state machine usually translates into deeper interplay with the game.
In my PhD thesis, I have observed that a strong design for the basic elements of a virtual execution engine allows a much more efficient handling of tasks. Reading open-sourced code from other games conforted me in the idea that a plain 'think()' method for handling all the possible cases of a complex character is a nightmare to work with. In an interpreted language, it would translate into repeated evaluation of tons of conditionals that could otherwise have been avoided.
As a result, the line between compiled and scripted code revolves around those state machines: the compiled code executes whatever is to happen in the current state, a data structure (built from script parsing) defines links between states, and interpreted expressions rule conditional transitions and side-effects, which allows strong reduction of the amount of interpreted symbols per game frame.
this is the right place for quickstuff
No comments:
Post a Comment